The principal aim of this proposal is to further development of new methods for analyzing observational data bases and randomized trials of HIV-infected persons and the application of these methods to data obtained in randomized and observational studies in an attempt to help answer important open substantive questions concerning the treatment and course of HlV-related disease. The proposed approaches are based either on (i) the estimation of new classes of causal models which include structural nested models, marginal structural models (MSMs), direct effect structural nested models, continuous time structural nested models, and optimail regime structural models (SNMs). Many of the new methods are fundamentally epidemiologic in that they require data on time-dependent confounding factors, that is, risk factors for outcomes that also predict subsequent treatment with the drug or cofactor under study. In particular, we plan to further develop optimal regime SNMs and dynamic MSMs to help detemnine the optimal times to start HAART therapy and to change HAART regimens as a function of a subject's CD4 count, HIV RNA, clinical history, and, where available, results of genot^lc or phenotypic resistance testing. Our methods will be developed with the goal of directing analyzes and reanalyzes, with collaborators, of data from the HIV Causal Colioboration at HSPH . the Multicenler AIDS Cohort Study, The Women's Interagency HIV Study, The Swiss HIV Cohort Study, The Study of The Consequences of Protease Inhibitor Era (SCOPE), Pediatric Late Outcomes Protocol (PACTG 219) and the ALLRT study. RELEVANCE (See instructions): Observational methods are used to answer pressing causal questions that cannot be or have not yet been studied in randomized trials. In particular we are developing methods that are the best available to determine the optimal CD4 and HIV RNA levels at which to initiate HAAART therapy in HIV infected subjects and the optimal time to change therapy once resistance to a initial HAART regime has developed.